Message-ID: <YvWdcX1Beo9ZbFXh@google.com>
Date:   Fri, 12 Aug 2022 10:23:13 +1000
From:   Matthew Bobrowski <repnop@...gle.com>
To:     Richard Guy Briggs <rgb@...hat.com>
Cc:     Linux-Audit Mailing List <linux-audit@...hat.com>,
        LKML <linux-kernel@...r.kernel.org>,
        linux-fsdevel@...r.kernel.org, Paul Moore <paul@...l-moore.com>,
        Eric Paris <eparis@...isplace.org>,
        Steve Grubb <sgrubb@...hat.com>, Jan Kara <jack@...e.cz>,
        Amir Goldstein <amir73il@...il.com>
Subject: Re: [PATCH v4 2/4] fanotify: define struct members to hold response
 decision context

On Tue, Aug 09, 2022 at 01:22:53PM -0400, Richard Guy Briggs wrote:
> This patch adds a flag, FAN_INFO and an extensible buffer to provide
> additional information about response decisions.  The buffer contains
> one or more headers defining the information type and the length of the
> following information.  The patch defines one additional information
> type, FAN_RESPONSE_INFO_AUDIT_RULE, an audit rule number.  This will
> allow for the creation of other information types in the future if other
> users of the API identify different needs.
> 
> Suggested-by: Steve Grubb <sgrubb@...hat.com>
> Link: https://lore.kernel.org/r/2745105.e9J7NaK4W3@x2
> Suggested-by: Jan Kara <jack@...e.cz>
> Link: https://lore.kernel.org/r/20201001101219.GE17860@quack2.suse.cz
> Signed-off-by: Richard Guy Briggs <rgb@...hat.com>
> ---

Just some comments/questions.

>  fs/notify/fanotify/fanotify.c      |  10 ++-
>  fs/notify/fanotify/fanotify.h      |   2 +
>  fs/notify/fanotify/fanotify_user.c | 104 +++++++++++++++++++++++------
>  include/linux/fanotify.h           |   5 ++
>  include/uapi/linux/fanotify.h      |  27 +++++++-
>  5 files changed, 123 insertions(+), 25 deletions(-)
> 
> diff --git a/fs/notify/fanotify/fanotify.c b/fs/notify/fanotify/fanotify.c
> index 4f897e109547..0f36062521f4 100644
> --- a/fs/notify/fanotify/fanotify.c
> +++ b/fs/notify/fanotify/fanotify.c
> @@ -262,13 +262,16 @@ static int fanotify_get_response(struct fsnotify_group *group,
>  	}
>  
>  	/* userspace responded, convert to something usable */
> -	switch (event->response & ~FAN_AUDIT) {
> +	switch (event->response & FANOTIFY_RESPONSE_ACCESS) {
>  	case FAN_ALLOW:
>  		ret = 0;
>  		break;
>  	case FAN_DENY:
> -	default:
>  		ret = -EPERM;
> +		break;
> +	default:
> +		ret = -EINVAL;
> +		break;

I'm definitely of the opinion that this shouldn't change as it
completely misrepresents why the acting process failed to perform
whatever operation it is intending to perform on the underlying
file.

Also, at this point, will we ever get into a situation where the
permission event was responded to using an invalid access response
flag? That is, if the event listener responds to a permission event
with something other than FAN_{ALLOW, DENY}, then it in turn already
receives an -EINVAL error. Having said that, I don't see how we'd ever
get to the point where the access list would contain a queued
permission event with an invalid response flag. I'm not saying that
this check should be dropped entirely, but rather returning -EINVAL
AFAICT to the actor process really doesn't make sense.

>  	/* Check if the response should be audited */
> @@ -560,6 +563,8 @@ static struct fanotify_event *fanotify_alloc_perm_event(const struct path *path,
>  
>  	pevent->fae.type = FANOTIFY_EVENT_TYPE_PATH_PERM;
>  	pevent->response = 0;
> +	pevent->info_len = 0;
> +	pevent->info_buf = NULL;
>  	pevent->state = FAN_EVENT_INIT;
>  	pevent->path = *path;
>  	path_get(path);
> @@ -996,6 +1001,7 @@ static void fanotify_free_path_event(struct fanotify_event *event)
>  static void fanotify_free_perm_event(struct fanotify_event *event)
>  {
>  	path_put(fanotify_event_path(event));
> +	kfree(FANOTIFY_PERM(event)->info_buf);
>  	kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
>  }
>  
> diff --git a/fs/notify/fanotify/fanotify.h b/fs/notify/fanotify/fanotify.h
> index abfa3712c185..14c30e173632 100644
> --- a/fs/notify/fanotify/fanotify.h
> +++ b/fs/notify/fanotify/fanotify.h
> @@ -428,6 +428,8 @@ struct fanotify_perm_event {
>  	u32 response;			/* userspace answer to the event */
>  	unsigned short state;		/* state of the event */
>  	int fd;		/* fd we passed to userspace for this event */
> +	size_t info_len;
> +	char *info_buf;
>  };
>  
>  static inline struct fanotify_perm_event *
> diff --git a/fs/notify/fanotify/fanotify_user.c b/fs/notify/fanotify/fanotify_user.c
> index ff67ca0d25cc..a4ae953f0e62 100644
> --- a/fs/notify/fanotify/fanotify_user.c
> +++ b/fs/notify/fanotify/fanotify_user.c
> @@ -289,13 +289,18 @@ static int create_fd(struct fsnotify_group *group, struct path *path,
>   */
>  static void finish_permission_event(struct fsnotify_group *group,
>  				    struct fanotify_perm_event *event,
> -				    u32 response)
> +				    struct fanotify_response *response,
> +				    size_t info_len, char *info_buf)
>  				    __releases(&group->notification_lock)
>  {
>  	bool destroy = false;
>  
>  	assert_spin_locked(&group->notification_lock);
> -	event->response = response;
> +	event->response = response->response & ~FAN_INFO;
> +	if (response->response & FAN_INFO) {
> +		event->info_len = info_len;
> +		event->info_buf = info_buf;
> +	}
>  	if (event->state == FAN_EVENT_CANCELED)
>  		destroy = true;
>  	else
> @@ -306,33 +311,71 @@ static void finish_permission_event(struct fsnotify_group *group,
>  }
>  
>  static int process_access_response(struct fsnotify_group *group,
> -				   struct fanotify_response *response_struct)
> +				   struct fanotify_response *response_struct,
> +				   const char __user *buf,
> +				   size_t count)
>  {
>  	struct fanotify_perm_event *event;
>  	int fd = response_struct->fd;
>  	u32 response = response_struct->response;
> +	struct fanotify_response_info_header info_hdr;

Why is this scoped at the function level?

> +	char *info_buf = NULL;
>  
> -	pr_debug("%s: group=%p fd=%d response=%u\n", __func__, group,
> -		 fd, response);
> +	pr_debug("%s: group=%p fd=%d response=%u buf=%p size=%lu\n", __func__,
> +		 group, fd, response, info_buf, count);

info_buf is NULL at this point, I think you meant buf,right? Also
s/size/count in the format string.

>  	/*
>  	 * make sure the response is valid, if invalid we do nothing and either
>  	 * userspace can send a valid response or we will clean it up after the
>  	 * timeout
>  	 */
> -	switch (response & ~FAN_AUDIT) {
> +	if (response & ~FANOTIFY_RESPONSE_VALID_MASK)
> +		return -EINVAL;
> +	switch (response & FANOTIFY_RESPONSE_ACCESS) {
>  	case FAN_ALLOW:
>  	case FAN_DENY:
>  		break;
>  	default:
>  		return -EINVAL;
>  	}
> -
> -	if (fd < 0)
> -		return -EINVAL;
> -
>  	if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT))
>  		return -EINVAL;
> +	if (fd < 0)
> +		return -EINVAL;
> +	if (response & FAN_INFO) {
> +		size_t c = count;
> +		const char __user *ib = buf;

Again, can we use variable names that are a little bit more
meaningful? *ib specifically confuses me as I've already got *info_buf
from above in my head. Maybe *info would make more sense seeing as
though this is the FAN_INFO path?

> +		If (c <= 0)
> +			return -EINVAL;

Is this needed? We already perform checks on the supplied count in
fanotify_write()?

> +		while (c >= sizeof(info_hdr)) {
> +			if (copy_from_user(&info_hdr, ib, sizeof(info_hdr)))
> +		 		return -EFAULT;
> +			if (info_hdr.pad != 0)
> +				return -EINVAL;
> +			if (c < info_hdr.len)
> +				return -EINVAL;
> +			switch (info_hdr.type) {
> +			case FAN_RESPONSE_INFO_AUDIT_RULE:
> +				break;
> +			case FAN_RESPONSE_INFO_NONE:
> +			default:
> +				return -EINVAL;
> +			}
> +			c -= info_hdr.len;
> +			ib += info_hdr.len;
> +		}
> +		if (c != 0)
> +			return -EINVAL;
> +		/* Simplistic check for now */
> +		if (count != sizeof(struct fanotify_response_info_audit_rule))
> +			return -EINVAL;

I don't get why we perform this check here? If anything, I'd expect
this to be one of the first things we do when we step into this
branch. There's no point of pulling the info_hdr if count isn't what
we expect?

> +		info_buf = kmalloc(sizeof(struct fanotify_response_info_audit_rule),
> +				   GFP_KERNEL);
> +		if (!info_buf)
> +			return -ENOMEM;
> +		if (copy_from_user(info_buf, buf, count))
> +			return -EFAULT;
> +	}

MY. EYES. HURT! This block is rather difficult to read, so feel free
to add newlines when splitting this up into a helper.

>  	spin_lock(&group->notification_lock);
>  	list_for_each_entry(event, &group->fanotify_data.access_list,
>  			    fae.fse.list) {
> @@ -340,7 +383,9 @@ static int process_access_response(struct fsnotify_group *group,
>  			continue;
>  
>  		list_del_init(&event->fae.fse.list);
> -		finish_permission_event(group, event, response);
> +		/* finish_permission_event() eats info_buf */

What is this comment? Get rid of it.

> +		finish_permission_event(group, event, response_struct,
> +					count, info_buf);
>  		wake_up(&group->fanotify_data.access_waitq);
>  		return 0;
>  	}
> @@ -802,9 +847,14 @@ static ssize_t fanotify_read(struct file *file, char __user *buf,
>  			fsnotify_destroy_event(group, &event->fse);
>  		} else {
>  			if (ret <= 0) {
> +				struct fanotify_response response = {
> +					.fd = FAN_NOFD,
> +					.response = FAN_DENY };
> +
>  				spin_lock(&group->notification_lock);
>  				finish_permission_event(group,
> -					FANOTIFY_PERM(event), FAN_DENY);
> +					FANOTIFY_PERM(event), &response,
> +					0, NULL);
>  				wake_up(&group->fanotify_data.access_waitq);
>  			} else {
>  				spin_lock(&group->notification_lock);
> @@ -827,26 +877,33 @@ static ssize_t fanotify_read(struct file *file, char __user *buf,
>  
>  static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
>  {
> -	struct fanotify_response response = { .fd = -1, .response = -1 };
> +	struct fanotify_response response;
>  	struct fsnotify_group *group;
>  	int ret;
> +	const char __user *info_buf = buf + sizeof(struct fanotify_response);
> +	size_t c;

Can we rename this to something like len or info_len instead? I
dislike single character variable names outside of the scope of things
like loops.

>  	if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS))
>  		return -EINVAL;
>  
>  	group = file->private_data;
>  
> -	if (count < sizeof(response))
> -		return -EINVAL;
> -
> -	count = sizeof(response);
> -
>  	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
>  
> -	if (copy_from_user(&response, buf, count))
> +	if (count < sizeof(response))
> +		return -EINVAL;
> +	if (copy_from_user(&response, buf, sizeof(response)))
>  		return -EFAULT;
>  
> -	ret = process_access_response(group, &response);
> +	c = count - sizeof(response);
> +	if (response.response & FAN_INFO) {
> +		if (c < sizeof(struct fanotify_response_info_header))
> +			return -EINVAL;
> +	} else {
> +		if (c != 0)
> +			return -EINVAL;

Hm, prior to this change we truncated the copy operation to the
sizeof(struct fanotify_response) and didn't care if there maybe was
extra data supplied in the buf or count > sizeof(struct
fanotify_response). This leaves me wondering whether this check is
needed for cases that are not (FAN_INFO | FAN_AUDIT)? The buf may
still hold a valid fanotify_response despite buf/count possibly being
larger than sizeof(struct fanotify_response)... I can see why you'd
want to enforce this, but I'm wondering if it might break things if
event listeners are responding to the permission events in an awkward
way i.e. by calculating and supplying count incorrectly.

Also, if we do decide to keep this check around, then maybe it can be
simplified into an else if instead?

> +	}
> +	ret = process_access_response(group, &response, info_buf, c);

Can we add a newline above this call to process_access_response()?

>  	if (ret < 0)
>  		count = ret;
>  
> @@ -857,6 +914,9 @@ static int fanotify_release(struct inode *ignored, struct file *file)
>  {
>  	struct fsnotify_group *group = file->private_data;
>  	struct fsnotify_event *fsn_event;
> +	struct fanotify_response response = {
> +		.fd = FAN_NOFD,
> +		.response = FAN_ALLOW };
>  
>  	/*
>  	 * Stop new events from arriving in the notification queue. since
> @@ -876,7 +936,7 @@ static int fanotify_release(struct inode *ignored, struct file *file)
>  		event = list_first_entry(&group->fanotify_data.access_list,
>  				struct fanotify_perm_event, fae.fse.list);
>  		list_del_init(&event->fae.fse.list);
> -		finish_permission_event(group, event, FAN_ALLOW);
> +		finish_permission_event(group, event, &response, 0, NULL);
>  		spin_lock(&group->notification_lock);
>  	}
>  
> @@ -893,7 +953,7 @@ static int fanotify_release(struct inode *ignored, struct file *file)
>  			fsnotify_destroy_event(group, fsn_event);
>  		} else {
>  			finish_permission_event(group, FANOTIFY_PERM(event),
> -						FAN_ALLOW);
> +						&response, 0, NULL);
>  		}
>  		spin_lock(&group->notification_lock);
>  	}
> diff --git a/include/linux/fanotify.h b/include/linux/fanotify.h
> index edc28555814c..ce9f97eb69f2 100644
> --- a/include/linux/fanotify.h
> +++ b/include/linux/fanotify.h
> @@ -114,6 +114,11 @@
>  #define ALL_FANOTIFY_EVENT_BITS		(FANOTIFY_OUTGOING_EVENTS | \
>  					 FANOTIFY_EVENT_FLAGS)
>  
> +/* This mask is to check for invalid bits of a user space permission response */

These masks are used across checks which involve permission responses.

> +#define FANOTIFY_RESPONSE_ACCESS (FAN_ALLOW | FAN_DENY)
> +#define FANOTIFY_RESPONSE_FLAGS (FAN_AUDIT | FAN_INFO)
> +#define FANOTIFY_RESPONSE_VALID_MASK (FANOTIFY_RESPONSE_ACCESS | FANOTIFY_RESPONSE_FLAGS)
> +
>  /* Do not use these old uapi constants internally */
>  #undef FAN_ALL_CLASS_BITS
>  #undef FAN_ALL_INIT_FLAGS
> diff --git a/include/uapi/linux/fanotify.h b/include/uapi/linux/fanotify.h
> index f1f89132d60e..4d08823a5698 100644
> --- a/include/uapi/linux/fanotify.h
> +++ b/include/uapi/linux/fanotify.h
> @@ -180,15 +180,40 @@ struct fanotify_event_info_error {
>  	__u32 error_count;
>  };
>  
> +/*
> + * User space may need to record additional information about its decision.
> + * The extra information type records what kind of information is included.
> + * The default is none. We also define an extra information buffer whose
> + * size is determined by the extra information type.
> + *
> + * If the context type is Rule, then the context following is the rule number
> + * that triggered the user space decision.

I'm actually confused by this last paragraph. What is "context type"
and what is "Rule"? Do you mean the struct
fanotify_response_info_header.type and the audit_rule that follows?

> +#define FAN_RESPONSE_INFO_NONE		0
> +#define FAN_RESPONSE_INFO_AUDIT_RULE	1
> +
>  struct fanotify_response {
>  	__s32 fd;
>  	__u32 response;
>  };
>  
> +struct fanotify_response_info_header {
> +	__u8 type;
> +	__u8 pad;
> +	__u16 len;
> +};
> +
> +struct fanotify_response_info_audit_rule {
> +	struct fanotify_response_info_header hdr;
> +	__u32 audit_rule;
> +};
> +
>  /* Legit userspace responses to a _PERM event */
>  #define FAN_ALLOW	0x01
>  #define FAN_DENY	0x02
> -#define FAN_AUDIT	0x10	/* Bit mask to create audit record for result */
> +#define FAN_AUDIT	0x10	/* Bitmask to create audit record for result */
> +#define FAN_INFO	0x20	/* Bitmask to indicate additional information */
>  
>  /* No fd set in event */
>  #define FAN_NOFD	-1
> -- 
> 2.27.0
> 

/M

Powered by blists - more mailing lists